2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
28 * Cookies are numeric values sent with CHANGE and REMOVE
29 * uevents while resuming, removing or renaming the device.
31 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
32 #define DM_COOKIE_LENGTH 24
34 static const char *_name
= DM_NAME
;
36 static unsigned int major
= 0;
37 static unsigned int _major
= 0;
39 static DEFINE_SPINLOCK(_minor_lock
);
42 * One of these is allocated per bio.
45 struct mapped_device
*md
;
49 unsigned long start_time
;
50 spinlock_t endio_lock
;
55 * One of these is allocated per target within a bio. Hopefully
56 * this will be simplified out one day.
65 * For request-based dm.
66 * One of these is allocated per request.
68 struct dm_rq_target_io
{
69 struct mapped_device
*md
;
71 struct request
*orig
, clone
;
77 * For request-based dm.
78 * One of these is allocated per bio.
80 struct dm_rq_clone_bio_info
{
82 struct dm_rq_target_io
*tio
;
85 union map_info
*dm_get_mapinfo(struct bio
*bio
)
87 if (bio
&& bio
->bi_private
)
88 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
92 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
94 if (rq
&& rq
->end_io_data
)
95 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
98 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
100 #define MINOR_ALLOCED ((void *)-1)
103 * Bits for the md->flags field.
105 #define DMF_BLOCK_IO_FOR_SUSPEND 0
106 #define DMF_SUSPENDED 1
108 #define DMF_FREEING 3
109 #define DMF_DELETING 4
110 #define DMF_NOFLUSH_SUSPENDING 5
111 #define DMF_QUEUE_IO_TO_THREAD 6
114 * Work processed by per-device workqueue.
116 struct mapped_device
{
117 struct rw_semaphore io_lock
;
118 struct mutex suspend_lock
;
125 struct request_queue
*queue
;
126 struct gendisk
*disk
;
132 * A list of ios that arrived while we were suspended.
135 wait_queue_head_t wait
;
136 struct work_struct work
;
137 struct bio_list deferred
;
138 spinlock_t deferred_lock
;
141 * An error from the barrier request currently being processed.
146 * Protect barrier_error from concurrent endio processing
147 * in request-based dm.
149 spinlock_t barrier_error_lock
;
152 * Processing queue (flush/barriers)
154 struct workqueue_struct
*wq
;
155 struct work_struct barrier_work
;
157 /* A pointer to the currently processing pre/post flush request */
158 struct request
*flush_request
;
161 * The current mapping.
163 struct dm_table
*map
;
166 * io objects are allocated from here.
177 wait_queue_head_t eventq
;
179 struct list_head uevent_list
;
180 spinlock_t uevent_lock
; /* Protect access to uevent_list */
183 * freeze/thaw support require holding onto a super block
185 struct super_block
*frozen_sb
;
186 struct block_device
*bdev
;
188 /* forced geometry settings */
189 struct hd_geometry geometry
;
191 /* For saving the address of __make_request for request based dm */
192 make_request_fn
*saved_make_request_fn
;
197 /* zero-length barrier that will be cloned and submitted to targets */
198 struct bio barrier_bio
;
202 * For mempools pre-allocation at the table loading time.
204 struct dm_md_mempools
{
211 static struct kmem_cache
*_io_cache
;
212 static struct kmem_cache
*_tio_cache
;
213 static struct kmem_cache
*_rq_tio_cache
;
214 static struct kmem_cache
*_rq_bio_info_cache
;
216 static int __init
local_init(void)
220 /* allocate a slab for the dm_ios */
221 _io_cache
= KMEM_CACHE(dm_io
, 0);
225 /* allocate a slab for the target ios */
226 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
228 goto out_free_io_cache
;
230 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
232 goto out_free_tio_cache
;
234 _rq_bio_info_cache
= KMEM_CACHE(dm_rq_clone_bio_info
, 0);
235 if (!_rq_bio_info_cache
)
236 goto out_free_rq_tio_cache
;
238 r
= dm_uevent_init();
240 goto out_free_rq_bio_info_cache
;
243 r
= register_blkdev(_major
, _name
);
245 goto out_uevent_exit
;
254 out_free_rq_bio_info_cache
:
255 kmem_cache_destroy(_rq_bio_info_cache
);
256 out_free_rq_tio_cache
:
257 kmem_cache_destroy(_rq_tio_cache
);
259 kmem_cache_destroy(_tio_cache
);
261 kmem_cache_destroy(_io_cache
);
266 static void local_exit(void)
268 kmem_cache_destroy(_rq_bio_info_cache
);
269 kmem_cache_destroy(_rq_tio_cache
);
270 kmem_cache_destroy(_tio_cache
);
271 kmem_cache_destroy(_io_cache
);
272 unregister_blkdev(_major
, _name
);
277 DMINFO("cleaned up");
280 static int (*_inits
[])(void) __initdata
= {
290 static void (*_exits
[])(void) = {
300 static int __init
dm_init(void)
302 const int count
= ARRAY_SIZE(_inits
);
306 for (i
= 0; i
< count
; i
++) {
321 static void __exit
dm_exit(void)
323 int i
= ARRAY_SIZE(_exits
);
330 * Block device functions
332 int dm_deleting_md(struct mapped_device
*md
)
334 return test_bit(DMF_DELETING
, &md
->flags
);
337 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
339 struct mapped_device
*md
;
341 spin_lock(&_minor_lock
);
343 md
= bdev
->bd_disk
->private_data
;
347 if (test_bit(DMF_FREEING
, &md
->flags
) ||
348 dm_deleting_md(md
)) {
354 atomic_inc(&md
->open_count
);
357 spin_unlock(&_minor_lock
);
359 return md
? 0 : -ENXIO
;
362 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
364 struct mapped_device
*md
= disk
->private_data
;
365 atomic_dec(&md
->open_count
);
370 int dm_open_count(struct mapped_device
*md
)
372 return atomic_read(&md
->open_count
);
376 * Guarantees nothing is using the device before it's deleted.
378 int dm_lock_for_deletion(struct mapped_device
*md
)
382 spin_lock(&_minor_lock
);
384 if (dm_open_count(md
))
387 set_bit(DMF_DELETING
, &md
->flags
);
389 spin_unlock(&_minor_lock
);
394 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
396 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
398 return dm_get_geometry(md
, geo
);
401 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
402 unsigned int cmd
, unsigned long arg
)
404 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
405 struct dm_table
*map
= dm_get_live_table(md
);
406 struct dm_target
*tgt
;
409 if (!map
|| !dm_table_get_size(map
))
412 /* We only support devices that have a single target */
413 if (dm_table_get_num_targets(map
) != 1)
416 tgt
= dm_table_get_target(map
, 0);
418 if (dm_suspended_md(md
)) {
423 if (tgt
->type
->ioctl
)
424 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
432 static struct dm_io
*alloc_io(struct mapped_device
*md
)
434 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
437 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
439 mempool_free(io
, md
->io_pool
);
442 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
444 mempool_free(tio
, md
->tio_pool
);
447 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
450 return mempool_alloc(md
->tio_pool
, gfp_mask
);
453 static void free_rq_tio(struct dm_rq_target_io
*tio
)
455 mempool_free(tio
, tio
->md
->tio_pool
);
458 static struct dm_rq_clone_bio_info
*alloc_bio_info(struct mapped_device
*md
)
460 return mempool_alloc(md
->io_pool
, GFP_ATOMIC
);
463 static void free_bio_info(struct dm_rq_clone_bio_info
*info
)
465 mempool_free(info
, info
->tio
->md
->io_pool
);
468 static int md_in_flight(struct mapped_device
*md
)
470 return atomic_read(&md
->pending
[READ
]) +
471 atomic_read(&md
->pending
[WRITE
]);
474 static void start_io_acct(struct dm_io
*io
)
476 struct mapped_device
*md
= io
->md
;
478 int rw
= bio_data_dir(io
->bio
);
480 io
->start_time
= jiffies
;
482 cpu
= part_stat_lock();
483 part_round_stats(cpu
, &dm_disk(md
)->part0
);
485 dm_disk(md
)->part0
.in_flight
[rw
] = atomic_inc_return(&md
->pending
[rw
]);
488 static void end_io_acct(struct dm_io
*io
)
490 struct mapped_device
*md
= io
->md
;
491 struct bio
*bio
= io
->bio
;
492 unsigned long duration
= jiffies
- io
->start_time
;
494 int rw
= bio_data_dir(bio
);
496 cpu
= part_stat_lock();
497 part_round_stats(cpu
, &dm_disk(md
)->part0
);
498 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
502 * After this is decremented the bio must not be touched if it is
505 dm_disk(md
)->part0
.in_flight
[rw
] = pending
=
506 atomic_dec_return(&md
->pending
[rw
]);
507 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
509 /* nudge anyone waiting on suspend queue */
515 * Add the bio to the list of deferred io.
517 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
519 down_write(&md
->io_lock
);
521 spin_lock_irq(&md
->deferred_lock
);
522 bio_list_add(&md
->deferred
, bio
);
523 spin_unlock_irq(&md
->deferred_lock
);
525 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD
, &md
->flags
))
526 queue_work(md
->wq
, &md
->work
);
528 up_write(&md
->io_lock
);
532 * Everyone (including functions in this file), should use this
533 * function to access the md->map field, and make sure they call
534 * dm_table_put() when finished.
536 struct dm_table
*dm_get_live_table(struct mapped_device
*md
)
541 read_lock_irqsave(&md
->map_lock
, flags
);
545 read_unlock_irqrestore(&md
->map_lock
, flags
);
551 * Get the geometry associated with a dm device
553 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
561 * Set the geometry of a device.
563 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
565 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
567 if (geo
->start
> sz
) {
568 DMWARN("Start sector is beyond the geometry limits.");
577 /*-----------------------------------------------------------------
579 * A more elegant soln is in the works that uses the queue
580 * merge fn, unfortunately there are a couple of changes to
581 * the block layer that I want to make for this. So in the
582 * interests of getting something for people to use I give
583 * you this clearly demarcated crap.
584 *---------------------------------------------------------------*/
586 static int __noflush_suspending(struct mapped_device
*md
)
588 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
592 * Decrements the number of outstanding ios that a bio has been
593 * cloned into, completing the original io if necc.
595 static void dec_pending(struct dm_io
*io
, int error
)
600 struct mapped_device
*md
= io
->md
;
602 /* Push-back supersedes any I/O errors */
603 if (unlikely(error
)) {
604 spin_lock_irqsave(&io
->endio_lock
, flags
);
605 if (!(io
->error
> 0 && __noflush_suspending(md
)))
607 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
610 if (atomic_dec_and_test(&io
->io_count
)) {
611 if (io
->error
== DM_ENDIO_REQUEUE
) {
613 * Target requested pushing back the I/O.
615 spin_lock_irqsave(&md
->deferred_lock
, flags
);
616 if (__noflush_suspending(md
)) {
617 if (!bio_rw_flagged(io
->bio
, BIO_RW_BARRIER
))
618 bio_list_add_head(&md
->deferred
,
621 /* noflush suspend was interrupted. */
623 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
626 io_error
= io
->error
;
629 if (bio_rw_flagged(bio
, BIO_RW_BARRIER
)) {
631 * There can be just one barrier request so we use
632 * a per-device variable for error reporting.
633 * Note that you can't touch the bio after end_io_acct
635 if (!md
->barrier_error
&& io_error
!= -EOPNOTSUPP
)
636 md
->barrier_error
= io_error
;
641 if (io_error
!= DM_ENDIO_REQUEUE
) {
642 trace_block_bio_complete(md
->queue
, bio
);
644 bio_endio(bio
, io_error
);
652 static void clone_endio(struct bio
*bio
, int error
)
655 struct dm_target_io
*tio
= bio
->bi_private
;
656 struct dm_io
*io
= tio
->io
;
657 struct mapped_device
*md
= tio
->io
->md
;
658 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
660 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
664 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
665 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
667 * error and requeue request are handled
671 else if (r
== DM_ENDIO_INCOMPLETE
)
672 /* The target will handle the io */
675 DMWARN("unimplemented target endio return value: %d", r
);
681 * Store md for cleanup instead of tio which is about to get freed.
683 bio
->bi_private
= md
->bs
;
687 dec_pending(io
, error
);
691 * Partial completion handling for request-based dm
693 static void end_clone_bio(struct bio
*clone
, int error
)
695 struct dm_rq_clone_bio_info
*info
= clone
->bi_private
;
696 struct dm_rq_target_io
*tio
= info
->tio
;
697 struct bio
*bio
= info
->orig
;
698 unsigned int nr_bytes
= info
->orig
->bi_size
;
704 * An error has already been detected on the request.
705 * Once error occurred, just let clone->end_io() handle
711 * Don't notice the error to the upper layer yet.
712 * The error handling decision is made by the target driver,
713 * when the request is completed.
720 * I/O for the bio successfully completed.
721 * Notice the data completion to the upper layer.
725 * bios are processed from the head of the list.
726 * So the completing bio should always be rq->bio.
727 * If it's not, something wrong is happening.
729 if (tio
->orig
->bio
!= bio
)
730 DMERR("bio completion is going in the middle of the request");
733 * Update the original request.
734 * Do not use blk_end_request() here, because it may complete
735 * the original request before the clone, and break the ordering.
737 blk_update_request(tio
->orig
, 0, nr_bytes
);
740 static void store_barrier_error(struct mapped_device
*md
, int error
)
744 spin_lock_irqsave(&md
->barrier_error_lock
, flags
);
746 * Basically, the first error is taken, but:
747 * -EOPNOTSUPP supersedes any I/O error.
748 * Requeue request supersedes any I/O error but -EOPNOTSUPP.
750 if (!md
->barrier_error
|| error
== -EOPNOTSUPP
||
751 (md
->barrier_error
!= -EOPNOTSUPP
&&
752 error
== DM_ENDIO_REQUEUE
))
753 md
->barrier_error
= error
;
754 spin_unlock_irqrestore(&md
->barrier_error_lock
, flags
);
758 * Don't touch any member of the md after calling this function because
759 * the md may be freed in dm_put() at the end of this function.
760 * Or do dm_get() before calling this function and dm_put() later.
762 static void rq_completed(struct mapped_device
*md
, int rw
, int run_queue
)
764 atomic_dec(&md
->pending
[rw
]);
766 /* nudge anyone waiting on suspend queue */
767 if (!md_in_flight(md
))
771 blk_run_queue(md
->queue
);
774 * dm_put() must be at the end of this function. See the comment above
779 static void free_rq_clone(struct request
*clone
)
781 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
783 blk_rq_unprep_clone(clone
);
788 * Complete the clone and the original request.
789 * Must be called without queue lock.
791 static void dm_end_request(struct request
*clone
, int error
)
793 int rw
= rq_data_dir(clone
);
795 bool is_barrier
= blk_barrier_rq(clone
);
796 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
797 struct mapped_device
*md
= tio
->md
;
798 struct request
*rq
= tio
->orig
;
800 if (blk_pc_request(rq
) && !is_barrier
) {
801 rq
->errors
= clone
->errors
;
802 rq
->resid_len
= clone
->resid_len
;
806 * We are using the sense buffer of the original
808 * So setting the length of the sense data is enough.
810 rq
->sense_len
= clone
->sense_len
;
813 free_rq_clone(clone
);
815 if (unlikely(is_barrier
)) {
817 store_barrier_error(md
, error
);
820 blk_end_request_all(rq
, error
);
822 rq_completed(md
, rw
, run_queue
);
825 static void dm_unprep_request(struct request
*rq
)
827 struct request
*clone
= rq
->special
;
830 rq
->cmd_flags
&= ~REQ_DONTPREP
;
832 free_rq_clone(clone
);
836 * Requeue the original request of a clone.
838 void dm_requeue_unmapped_request(struct request
*clone
)
840 int rw
= rq_data_dir(clone
);
841 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
842 struct mapped_device
*md
= tio
->md
;
843 struct request
*rq
= tio
->orig
;
844 struct request_queue
*q
= rq
->q
;
847 if (unlikely(blk_barrier_rq(clone
))) {
849 * Barrier clones share an original request.
850 * Leave it to dm_end_request(), which handles this special
853 dm_end_request(clone
, DM_ENDIO_REQUEUE
);
857 dm_unprep_request(rq
);
859 spin_lock_irqsave(q
->queue_lock
, flags
);
860 if (elv_queue_empty(q
))
862 blk_requeue_request(q
, rq
);
863 spin_unlock_irqrestore(q
->queue_lock
, flags
);
865 rq_completed(md
, rw
, 0);
867 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request
);
869 static void __stop_queue(struct request_queue
*q
)
874 static void stop_queue(struct request_queue
*q
)
878 spin_lock_irqsave(q
->queue_lock
, flags
);
880 spin_unlock_irqrestore(q
->queue_lock
, flags
);
883 static void __start_queue(struct request_queue
*q
)
885 if (blk_queue_stopped(q
))
889 static void start_queue(struct request_queue
*q
)
893 spin_lock_irqsave(q
->queue_lock
, flags
);
895 spin_unlock_irqrestore(q
->queue_lock
, flags
);
898 static void dm_done(struct request
*clone
, int error
, bool mapped
)
901 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
902 dm_request_endio_fn rq_end_io
= tio
->ti
->type
->rq_end_io
;
904 if (mapped
&& rq_end_io
)
905 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
908 /* The target wants to complete the I/O */
909 dm_end_request(clone
, r
);
910 else if (r
== DM_ENDIO_INCOMPLETE
)
911 /* The target will handle the I/O */
913 else if (r
== DM_ENDIO_REQUEUE
)
914 /* The target wants to requeue the I/O */
915 dm_requeue_unmapped_request(clone
);
917 DMWARN("unimplemented target endio return value: %d", r
);
923 * Request completion handler for request-based dm
925 static void dm_softirq_done(struct request
*rq
)
928 struct request
*clone
= rq
->completion_data
;
929 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
931 if (rq
->cmd_flags
& REQ_FAILED
)
934 dm_done(clone
, tio
->error
, mapped
);
938 * Complete the clone and the original request with the error status
939 * through softirq context.
941 static void dm_complete_request(struct request
*clone
, int error
)
943 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
944 struct request
*rq
= tio
->orig
;
946 if (unlikely(blk_barrier_rq(clone
))) {
948 * Barrier clones share an original request. So can't use
949 * softirq_done with the original.
950 * Pass the clone to dm_done() directly in this special case.
951 * It is safe (even if clone->q->queue_lock is held here)
952 * because there is no I/O dispatching during the completion
955 dm_done(clone
, error
, true);
960 rq
->completion_data
= clone
;
961 blk_complete_request(rq
);
965 * Complete the not-mapped clone and the original request with the error status
966 * through softirq context.
967 * Target's rq_end_io() function isn't called.
968 * This may be used when the target's map_rq() function fails.
970 void dm_kill_unmapped_request(struct request
*clone
, int error
)
972 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
973 struct request
*rq
= tio
->orig
;
975 if (unlikely(blk_barrier_rq(clone
))) {
977 * Barrier clones share an original request.
978 * Leave it to dm_end_request(), which handles this special
982 dm_end_request(clone
, error
);
986 rq
->cmd_flags
|= REQ_FAILED
;
987 dm_complete_request(clone
, error
);
989 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request
);
992 * Called with the queue lock held
994 static void end_clone_request(struct request
*clone
, int error
)
997 * For just cleaning up the information of the queue in which
998 * the clone was dispatched.
999 * The clone is *NOT* freed actually here because it is alloced from
1000 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1002 __blk_put_request(clone
->q
, clone
);
1005 * Actual request completion is done in a softirq context which doesn't
1006 * hold the queue lock. Otherwise, deadlock could occur because:
1007 * - another request may be submitted by the upper level driver
1008 * of the stacking during the completion
1009 * - the submission which requires queue lock may be done
1010 * against this queue
1012 dm_complete_request(clone
, error
);
1015 static sector_t
max_io_len(struct mapped_device
*md
,
1016 sector_t sector
, struct dm_target
*ti
)
1018 sector_t offset
= sector
- ti
->begin
;
1019 sector_t len
= ti
->len
- offset
;
1022 * Does the target need to split even further ?
1026 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
1035 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
1036 struct dm_target_io
*tio
)
1040 struct mapped_device
*md
;
1042 clone
->bi_end_io
= clone_endio
;
1043 clone
->bi_private
= tio
;
1046 * Map the clone. If r == 0 we don't need to do
1047 * anything, the target has assumed ownership of
1050 atomic_inc(&tio
->io
->io_count
);
1051 sector
= clone
->bi_sector
;
1052 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
1053 if (r
== DM_MAPIO_REMAPPED
) {
1054 /* the bio has been remapped so dispatch it */
1056 trace_block_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1057 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1059 generic_make_request(clone
);
1060 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1061 /* error the io and bail out, or requeue it if needed */
1063 dec_pending(tio
->io
, r
);
1065 * Store bio_set for cleanup.
1067 clone
->bi_private
= md
->bs
;
1071 DMWARN("unimplemented target map return value: %d", r
);
1077 struct mapped_device
*md
;
1078 struct dm_table
*map
;
1082 sector_t sector_count
;
1086 static void dm_bio_destructor(struct bio
*bio
)
1088 struct bio_set
*bs
= bio
->bi_private
;
1094 * Creates a little bio that is just does part of a bvec.
1096 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
1097 unsigned short idx
, unsigned int offset
,
1098 unsigned int len
, struct bio_set
*bs
)
1101 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
1103 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
1104 clone
->bi_destructor
= dm_bio_destructor
;
1105 *clone
->bi_io_vec
= *bv
;
1107 clone
->bi_sector
= sector
;
1108 clone
->bi_bdev
= bio
->bi_bdev
;
1109 clone
->bi_rw
= bio
->bi_rw
& ~(1 << BIO_RW_BARRIER
);
1111 clone
->bi_size
= to_bytes(len
);
1112 clone
->bi_io_vec
->bv_offset
= offset
;
1113 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
1114 clone
->bi_flags
|= 1 << BIO_CLONED
;
1116 if (bio_integrity(bio
)) {
1117 bio_integrity_clone(clone
, bio
, GFP_NOIO
, bs
);
1118 bio_integrity_trim(clone
,
1119 bio_sector_offset(bio
, idx
, offset
), len
);
1126 * Creates a bio that consists of range of complete bvecs.
1128 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
1129 unsigned short idx
, unsigned short bv_count
,
1130 unsigned int len
, struct bio_set
*bs
)
1134 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
1135 __bio_clone(clone
, bio
);
1136 clone
->bi_rw
&= ~(1 << BIO_RW_BARRIER
);
1137 clone
->bi_destructor
= dm_bio_destructor
;
1138 clone
->bi_sector
= sector
;
1139 clone
->bi_idx
= idx
;
1140 clone
->bi_vcnt
= idx
+ bv_count
;
1141 clone
->bi_size
= to_bytes(len
);
1142 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
1144 if (bio_integrity(bio
)) {
1145 bio_integrity_clone(clone
, bio
, GFP_NOIO
, bs
);
1147 if (idx
!= bio
->bi_idx
|| clone
->bi_size
< bio
->bi_size
)
1148 bio_integrity_trim(clone
,
1149 bio_sector_offset(bio
, idx
, 0), len
);
1155 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1156 struct dm_target
*ti
)
1158 struct dm_target_io
*tio
= mempool_alloc(ci
->md
->tio_pool
, GFP_NOIO
);
1162 memset(&tio
->info
, 0, sizeof(tio
->info
));
1167 static void __flush_target(struct clone_info
*ci
, struct dm_target
*ti
,
1170 struct dm_target_io
*tio
= alloc_tio(ci
, ti
);
1173 tio
->info
.flush_request
= flush_nr
;
1175 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1176 __bio_clone(clone
, ci
->bio
);
1177 clone
->bi_destructor
= dm_bio_destructor
;
1179 __map_bio(ti
, clone
, tio
);
1182 static int __clone_and_map_empty_barrier(struct clone_info
*ci
)
1184 unsigned target_nr
= 0, flush_nr
;
1185 struct dm_target
*ti
;
1187 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1188 for (flush_nr
= 0; flush_nr
< ti
->num_flush_requests
;
1190 __flush_target(ci
, ti
, flush_nr
);
1192 ci
->sector_count
= 0;
1197 static int __clone_and_map(struct clone_info
*ci
)
1199 struct bio
*clone
, *bio
= ci
->bio
;
1200 struct dm_target
*ti
;
1201 sector_t len
= 0, max
;
1202 struct dm_target_io
*tio
;
1204 if (unlikely(bio_empty_barrier(bio
)))
1205 return __clone_and_map_empty_barrier(ci
);
1207 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1208 if (!dm_target_is_valid(ti
))
1211 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
1214 * Allocate a target io object.
1216 tio
= alloc_tio(ci
, ti
);
1218 if (ci
->sector_count
<= max
) {
1220 * Optimise for the simple case where we can do all of
1221 * the remaining io with a single clone.
1223 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
1224 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
1226 __map_bio(ti
, clone
, tio
);
1227 ci
->sector_count
= 0;
1229 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
1231 * There are some bvecs that don't span targets.
1232 * Do as many of these as possible.
1235 sector_t remaining
= max
;
1238 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
1239 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
1241 if (bv_len
> remaining
)
1244 remaining
-= bv_len
;
1248 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
1250 __map_bio(ti
, clone
, tio
);
1253 ci
->sector_count
-= len
;
1258 * Handle a bvec that must be split between two or more targets.
1260 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
1261 sector_t remaining
= to_sector(bv
->bv_len
);
1262 unsigned int offset
= 0;
1266 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1267 if (!dm_target_is_valid(ti
))
1270 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
1272 tio
= alloc_tio(ci
, ti
);
1275 len
= min(remaining
, max
);
1277 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
1278 bv
->bv_offset
+ offset
, len
,
1281 __map_bio(ti
, clone
, tio
);
1284 ci
->sector_count
-= len
;
1285 offset
+= to_bytes(len
);
1286 } while (remaining
-= len
);
1295 * Split the bio into several clones and submit it to targets.
1297 static void __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
1299 struct clone_info ci
;
1302 ci
.map
= dm_get_live_table(md
);
1303 if (unlikely(!ci
.map
)) {
1304 if (!bio_rw_flagged(bio
, BIO_RW_BARRIER
))
1307 if (!md
->barrier_error
)
1308 md
->barrier_error
= -EIO
;
1314 ci
.io
= alloc_io(md
);
1316 atomic_set(&ci
.io
->io_count
, 1);
1319 spin_lock_init(&ci
.io
->endio_lock
);
1320 ci
.sector
= bio
->bi_sector
;
1321 ci
.sector_count
= bio_sectors(bio
);
1322 if (unlikely(bio_empty_barrier(bio
)))
1323 ci
.sector_count
= 1;
1324 ci
.idx
= bio
->bi_idx
;
1326 start_io_acct(ci
.io
);
1327 while (ci
.sector_count
&& !error
)
1328 error
= __clone_and_map(&ci
);
1330 /* drop the extra reference count */
1331 dec_pending(ci
.io
, error
);
1332 dm_table_put(ci
.map
);
1334 /*-----------------------------------------------------------------
1336 *---------------------------------------------------------------*/
1338 static int dm_merge_bvec(struct request_queue
*q
,
1339 struct bvec_merge_data
*bvm
,
1340 struct bio_vec
*biovec
)
1342 struct mapped_device
*md
= q
->queuedata
;
1343 struct dm_table
*map
= dm_get_live_table(md
);
1344 struct dm_target
*ti
;
1345 sector_t max_sectors
;
1351 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1352 if (!dm_target_is_valid(ti
))
1356 * Find maximum amount of I/O that won't need splitting
1358 max_sectors
= min(max_io_len(md
, bvm
->bi_sector
, ti
),
1359 (sector_t
) BIO_MAX_SECTORS
);
1360 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1365 * merge_bvec_fn() returns number of bytes
1366 * it can accept at this offset
1367 * max is precomputed maximal io size
1369 if (max_size
&& ti
->type
->merge
)
1370 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1372 * If the target doesn't support merge method and some of the devices
1373 * provided their merge_bvec method (we know this by looking at
1374 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1375 * entries. So always set max_size to 0, and the code below allows
1378 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1387 * Always allow an entire first page
1389 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1390 max_size
= biovec
->bv_len
;
1396 * The request function that just remaps the bio built up by
1399 static int _dm_request(struct request_queue
*q
, struct bio
*bio
)
1401 int rw
= bio_data_dir(bio
);
1402 struct mapped_device
*md
= q
->queuedata
;
1405 down_read(&md
->io_lock
);
1407 cpu
= part_stat_lock();
1408 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
1409 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
1413 * If we're suspended or the thread is processing barriers
1414 * we have to queue this io for later.
1416 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD
, &md
->flags
)) ||
1417 unlikely(bio_rw_flagged(bio
, BIO_RW_BARRIER
))) {
1418 up_read(&md
->io_lock
);
1420 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) &&
1421 bio_rw(bio
) == READA
) {
1431 __split_and_process_bio(md
, bio
);
1432 up_read(&md
->io_lock
);
1436 static int dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1438 struct mapped_device
*md
= q
->queuedata
;
1440 return md
->saved_make_request_fn(q
, bio
); /* call __make_request() */
1443 static int dm_request_based(struct mapped_device
*md
)
1445 return blk_queue_stackable(md
->queue
);
1448 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
1450 struct mapped_device
*md
= q
->queuedata
;
1452 if (dm_request_based(md
))
1453 return dm_make_request(q
, bio
);
1455 return _dm_request(q
, bio
);
1459 * Mark this request as flush request, so that dm_request_fn() can
1462 static void dm_rq_prepare_flush(struct request_queue
*q
, struct request
*rq
)
1464 rq
->cmd_type
= REQ_TYPE_LINUX_BLOCK
;
1465 rq
->cmd
[0] = REQ_LB_OP_FLUSH
;
1468 static bool dm_rq_is_flush_request(struct request
*rq
)
1470 if (rq
->cmd_type
== REQ_TYPE_LINUX_BLOCK
&&
1471 rq
->cmd
[0] == REQ_LB_OP_FLUSH
)
1477 void dm_dispatch_request(struct request
*rq
)
1481 if (blk_queue_io_stat(rq
->q
))
1482 rq
->cmd_flags
|= REQ_IO_STAT
;
1484 rq
->start_time
= jiffies
;
1485 r
= blk_insert_cloned_request(rq
->q
, rq
);
1487 dm_complete_request(rq
, r
);
1489 EXPORT_SYMBOL_GPL(dm_dispatch_request
);
1491 static void dm_rq_bio_destructor(struct bio
*bio
)
1493 struct dm_rq_clone_bio_info
*info
= bio
->bi_private
;
1494 struct mapped_device
*md
= info
->tio
->md
;
1496 free_bio_info(info
);
1497 bio_free(bio
, md
->bs
);
1500 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1503 struct dm_rq_target_io
*tio
= data
;
1504 struct mapped_device
*md
= tio
->md
;
1505 struct dm_rq_clone_bio_info
*info
= alloc_bio_info(md
);
1510 info
->orig
= bio_orig
;
1512 bio
->bi_end_io
= end_clone_bio
;
1513 bio
->bi_private
= info
;
1514 bio
->bi_destructor
= dm_rq_bio_destructor
;
1519 static int setup_clone(struct request
*clone
, struct request
*rq
,
1520 struct dm_rq_target_io
*tio
)
1524 if (dm_rq_is_flush_request(rq
)) {
1525 blk_rq_init(NULL
, clone
);
1526 clone
->cmd_type
= REQ_TYPE_FS
;
1527 clone
->cmd_flags
|= (REQ_HARDBARRIER
| WRITE
);
1529 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, GFP_ATOMIC
,
1530 dm_rq_bio_constructor
, tio
);
1534 clone
->cmd
= rq
->cmd
;
1535 clone
->cmd_len
= rq
->cmd_len
;
1536 clone
->sense
= rq
->sense
;
1537 clone
->buffer
= rq
->buffer
;
1540 clone
->end_io
= end_clone_request
;
1541 clone
->end_io_data
= tio
;
1546 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1549 struct request
*clone
;
1550 struct dm_rq_target_io
*tio
;
1552 tio
= alloc_rq_tio(md
, gfp_mask
);
1560 memset(&tio
->info
, 0, sizeof(tio
->info
));
1562 clone
= &tio
->clone
;
1563 if (setup_clone(clone
, rq
, tio
)) {
1573 * Called with the queue lock held.
1575 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1577 struct mapped_device
*md
= q
->queuedata
;
1578 struct request
*clone
;
1580 if (unlikely(dm_rq_is_flush_request(rq
)))
1583 if (unlikely(rq
->special
)) {
1584 DMWARN("Already has something in rq->special.");
1585 return BLKPREP_KILL
;
1588 clone
= clone_rq(rq
, md
, GFP_ATOMIC
);
1590 return BLKPREP_DEFER
;
1592 rq
->special
= clone
;
1593 rq
->cmd_flags
|= REQ_DONTPREP
;
1598 static void map_request(struct dm_target
*ti
, struct request
*clone
,
1599 struct mapped_device
*md
)
1602 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1605 * Hold the md reference here for the in-flight I/O.
1606 * We can't rely on the reference count by device opener,
1607 * because the device may be closed during the request completion
1608 * when all bios are completed.
1609 * See the comment in rq_completed() too.
1614 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1616 case DM_MAPIO_SUBMITTED
:
1617 /* The target has taken the I/O to submit by itself later */
1619 case DM_MAPIO_REMAPPED
:
1620 /* The target has remapped the I/O so dispatch it */
1621 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1622 blk_rq_pos(tio
->orig
));
1623 dm_dispatch_request(clone
);
1625 case DM_MAPIO_REQUEUE
:
1626 /* The target wants to requeue the I/O */
1627 dm_requeue_unmapped_request(clone
);
1631 DMWARN("unimplemented target map return value: %d", r
);
1635 /* The target wants to complete the I/O */
1636 dm_kill_unmapped_request(clone
, r
);
1642 * q->request_fn for request-based dm.
1643 * Called with the queue lock held.
1645 static void dm_request_fn(struct request_queue
*q
)
1647 struct mapped_device
*md
= q
->queuedata
;
1648 struct dm_table
*map
= dm_get_live_table(md
);
1649 struct dm_target
*ti
;
1650 struct request
*rq
, *clone
;
1653 * For suspend, check blk_queue_stopped() and increment
1654 * ->pending within a single queue_lock not to increment the
1655 * number of in-flight I/Os after the queue is stopped in
1658 while (!blk_queue_plugged(q
) && !blk_queue_stopped(q
)) {
1659 rq
= blk_peek_request(q
);
1663 if (unlikely(dm_rq_is_flush_request(rq
))) {
1664 BUG_ON(md
->flush_request
);
1665 md
->flush_request
= rq
;
1666 blk_start_request(rq
);
1667 queue_work(md
->wq
, &md
->barrier_work
);
1671 ti
= dm_table_find_target(map
, blk_rq_pos(rq
));
1672 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1675 blk_start_request(rq
);
1676 clone
= rq
->special
;
1677 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
1679 spin_unlock(q
->queue_lock
);
1680 map_request(ti
, clone
, md
);
1681 spin_lock_irq(q
->queue_lock
);
1687 if (!elv_queue_empty(q
))
1688 /* Some requests still remain, retry later */
1697 int dm_underlying_device_busy(struct request_queue
*q
)
1699 return blk_lld_busy(q
);
1701 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
1703 static int dm_lld_busy(struct request_queue
*q
)
1706 struct mapped_device
*md
= q
->queuedata
;
1707 struct dm_table
*map
= dm_get_live_table(md
);
1709 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
1712 r
= dm_table_any_busy_target(map
);
1719 static void dm_unplug_all(struct request_queue
*q
)
1721 struct mapped_device
*md
= q
->queuedata
;
1722 struct dm_table
*map
= dm_get_live_table(md
);
1725 if (dm_request_based(md
))
1726 generic_unplug_device(q
);
1728 dm_table_unplug_all(map
);
1733 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1736 struct mapped_device
*md
= congested_data
;
1737 struct dm_table
*map
;
1739 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1740 map
= dm_get_live_table(md
);
1743 * Request-based dm cares about only own queue for
1744 * the query about congestion status of request_queue
1746 if (dm_request_based(md
))
1747 r
= md
->queue
->backing_dev_info
.state
&
1750 r
= dm_table_any_congested(map
, bdi_bits
);
1759 /*-----------------------------------------------------------------
1760 * An IDR is used to keep track of allocated minor numbers.
1761 *---------------------------------------------------------------*/
1762 static DEFINE_IDR(_minor_idr
);
1764 static void free_minor(int minor
)
1766 spin_lock(&_minor_lock
);
1767 idr_remove(&_minor_idr
, minor
);
1768 spin_unlock(&_minor_lock
);
1772 * See if the device with a specific minor # is free.
1774 static int specific_minor(int minor
)
1778 if (minor
>= (1 << MINORBITS
))
1781 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1785 spin_lock(&_minor_lock
);
1787 if (idr_find(&_minor_idr
, minor
)) {
1792 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1797 idr_remove(&_minor_idr
, m
);
1803 spin_unlock(&_minor_lock
);
1807 static int next_free_minor(int *minor
)
1811 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1815 spin_lock(&_minor_lock
);
1817 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1821 if (m
>= (1 << MINORBITS
)) {
1822 idr_remove(&_minor_idr
, m
);
1830 spin_unlock(&_minor_lock
);
1834 static const struct block_device_operations dm_blk_dops
;
1836 static void dm_wq_work(struct work_struct
*work
);
1837 static void dm_rq_barrier_work(struct work_struct
*work
);
1840 * Allocate and initialise a blank device with a given minor.
1842 static struct mapped_device
*alloc_dev(int minor
)
1845 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1849 DMWARN("unable to allocate device, out of memory.");
1853 if (!try_module_get(THIS_MODULE
))
1854 goto bad_module_get
;
1856 /* get a minor number for the dev */
1857 if (minor
== DM_ANY_MINOR
)
1858 r
= next_free_minor(&minor
);
1860 r
= specific_minor(minor
);
1864 init_rwsem(&md
->io_lock
);
1865 mutex_init(&md
->suspend_lock
);
1866 spin_lock_init(&md
->deferred_lock
);
1867 spin_lock_init(&md
->barrier_error_lock
);
1868 rwlock_init(&md
->map_lock
);
1869 atomic_set(&md
->holders
, 1);
1870 atomic_set(&md
->open_count
, 0);
1871 atomic_set(&md
->event_nr
, 0);
1872 atomic_set(&md
->uevent_seq
, 0);
1873 INIT_LIST_HEAD(&md
->uevent_list
);
1874 spin_lock_init(&md
->uevent_lock
);
1876 md
->queue
= blk_init_queue(dm_request_fn
, NULL
);
1881 * Request-based dm devices cannot be stacked on top of bio-based dm
1882 * devices. The type of this dm device has not been decided yet,
1883 * although we initialized the queue using blk_init_queue().
1884 * The type is decided at the first table loading time.
1885 * To prevent problematic device stacking, clear the queue flag
1886 * for request stacking support until then.
1888 * This queue is new, so no concurrency on the queue_flags.
1890 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1891 md
->saved_make_request_fn
= md
->queue
->make_request_fn
;
1892 md
->queue
->queuedata
= md
;
1893 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1894 md
->queue
->backing_dev_info
.congested_data
= md
;
1895 blk_queue_make_request(md
->queue
, dm_request
);
1896 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1897 md
->queue
->unplug_fn
= dm_unplug_all
;
1898 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1899 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
1900 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
1901 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
1902 blk_queue_ordered(md
->queue
, QUEUE_ORDERED_DRAIN_FLUSH
,
1903 dm_rq_prepare_flush
);
1905 md
->disk
= alloc_disk(1);
1909 atomic_set(&md
->pending
[0], 0);
1910 atomic_set(&md
->pending
[1], 0);
1911 init_waitqueue_head(&md
->wait
);
1912 INIT_WORK(&md
->work
, dm_wq_work
);
1913 INIT_WORK(&md
->barrier_work
, dm_rq_barrier_work
);
1914 init_waitqueue_head(&md
->eventq
);
1916 md
->disk
->major
= _major
;
1917 md
->disk
->first_minor
= minor
;
1918 md
->disk
->fops
= &dm_blk_dops
;
1919 md
->disk
->queue
= md
->queue
;
1920 md
->disk
->private_data
= md
;
1921 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1923 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1925 md
->wq
= create_singlethread_workqueue("kdmflush");
1929 md
->bdev
= bdget_disk(md
->disk
, 0);
1933 /* Populate the mapping, nobody knows we exist yet */
1934 spin_lock(&_minor_lock
);
1935 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1936 spin_unlock(&_minor_lock
);
1938 BUG_ON(old_md
!= MINOR_ALLOCED
);
1943 destroy_workqueue(md
->wq
);
1945 del_gendisk(md
->disk
);
1948 blk_cleanup_queue(md
->queue
);
1952 module_put(THIS_MODULE
);
1958 static void unlock_fs(struct mapped_device
*md
);
1960 static void free_dev(struct mapped_device
*md
)
1962 int minor
= MINOR(disk_devt(md
->disk
));
1966 destroy_workqueue(md
->wq
);
1968 mempool_destroy(md
->tio_pool
);
1970 mempool_destroy(md
->io_pool
);
1972 bioset_free(md
->bs
);
1973 blk_integrity_unregister(md
->disk
);
1974 del_gendisk(md
->disk
);
1977 spin_lock(&_minor_lock
);
1978 md
->disk
->private_data
= NULL
;
1979 spin_unlock(&_minor_lock
);
1982 blk_cleanup_queue(md
->queue
);
1983 module_put(THIS_MODULE
);
1987 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1989 struct dm_md_mempools
*p
;
1991 if (md
->io_pool
&& md
->tio_pool
&& md
->bs
)
1992 /* the md already has necessary mempools */
1995 p
= dm_table_get_md_mempools(t
);
1996 BUG_ON(!p
|| md
->io_pool
|| md
->tio_pool
|| md
->bs
);
1998 md
->io_pool
= p
->io_pool
;
2000 md
->tio_pool
= p
->tio_pool
;
2006 /* mempool bind completed, now no need any mempools in the table */
2007 dm_table_free_md_mempools(t
);
2011 * Bind a table to the device.
2013 static void event_callback(void *context
)
2015 unsigned long flags
;
2017 struct mapped_device
*md
= (struct mapped_device
*) context
;
2019 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2020 list_splice_init(&md
->uevent_list
, &uevents
);
2021 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2023 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2025 atomic_inc(&md
->event_nr
);
2026 wake_up(&md
->eventq
);
2029 static void __set_size(struct mapped_device
*md
, sector_t size
)
2031 set_capacity(md
->disk
, size
);
2033 mutex_lock(&md
->bdev
->bd_inode
->i_mutex
);
2034 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2035 mutex_unlock(&md
->bdev
->bd_inode
->i_mutex
);
2039 * Returns old map, which caller must destroy.
2041 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2042 struct queue_limits
*limits
)
2044 struct dm_table
*old_map
;
2045 struct request_queue
*q
= md
->queue
;
2047 unsigned long flags
;
2049 size
= dm_table_get_size(t
);
2052 * Wipe any geometry if the size of the table changed.
2054 if (size
!= get_capacity(md
->disk
))
2055 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2057 __set_size(md
, size
);
2059 dm_table_event_callback(t
, event_callback
, md
);
2062 * The queue hasn't been stopped yet, if the old table type wasn't
2063 * for request-based during suspension. So stop it to prevent
2064 * I/O mapping before resume.
2065 * This must be done before setting the queue restrictions,
2066 * because request-based dm may be run just after the setting.
2068 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2071 __bind_mempools(md
, t
);
2073 write_lock_irqsave(&md
->map_lock
, flags
);
2076 dm_table_set_restrictions(t
, q
, limits
);
2077 write_unlock_irqrestore(&md
->map_lock
, flags
);
2083 * Returns unbound table for the caller to free.
2085 static struct dm_table
*__unbind(struct mapped_device
*md
)
2087 struct dm_table
*map
= md
->map
;
2088 unsigned long flags
;
2093 dm_table_event_callback(map
, NULL
, NULL
);
2094 write_lock_irqsave(&md
->map_lock
, flags
);
2096 write_unlock_irqrestore(&md
->map_lock
, flags
);
2102 * Constructor for a new device.
2104 int dm_create(int minor
, struct mapped_device
**result
)
2106 struct mapped_device
*md
;
2108 md
= alloc_dev(minor
);
2118 static struct mapped_device
*dm_find_md(dev_t dev
)
2120 struct mapped_device
*md
;
2121 unsigned minor
= MINOR(dev
);
2123 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2126 spin_lock(&_minor_lock
);
2128 md
= idr_find(&_minor_idr
, minor
);
2129 if (md
&& (md
== MINOR_ALLOCED
||
2130 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2131 test_bit(DMF_FREEING
, &md
->flags
))) {
2137 spin_unlock(&_minor_lock
);
2142 struct mapped_device
*dm_get_md(dev_t dev
)
2144 struct mapped_device
*md
= dm_find_md(dev
);
2152 void *dm_get_mdptr(struct mapped_device
*md
)
2154 return md
->interface_ptr
;
2157 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2159 md
->interface_ptr
= ptr
;
2162 void dm_get(struct mapped_device
*md
)
2164 atomic_inc(&md
->holders
);
2167 const char *dm_device_name(struct mapped_device
*md
)
2171 EXPORT_SYMBOL_GPL(dm_device_name
);
2173 void dm_put(struct mapped_device
*md
)
2175 struct dm_table
*map
;
2177 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2179 if (atomic_dec_and_lock(&md
->holders
, &_minor_lock
)) {
2180 map
= dm_get_live_table(md
);
2181 idr_replace(&_minor_idr
, MINOR_ALLOCED
,
2182 MINOR(disk_devt(dm_disk(md
))));
2183 set_bit(DMF_FREEING
, &md
->flags
);
2184 spin_unlock(&_minor_lock
);
2185 if (!dm_suspended_md(md
)) {
2186 dm_table_presuspend_targets(map
);
2187 dm_table_postsuspend_targets(map
);
2191 dm_table_destroy(__unbind(md
));
2195 EXPORT_SYMBOL_GPL(dm_put
);
2197 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2200 DECLARE_WAITQUEUE(wait
, current
);
2202 dm_unplug_all(md
->queue
);
2204 add_wait_queue(&md
->wait
, &wait
);
2207 set_current_state(interruptible
);
2210 if (!md_in_flight(md
))
2213 if (interruptible
== TASK_INTERRUPTIBLE
&&
2214 signal_pending(current
)) {
2221 set_current_state(TASK_RUNNING
);
2223 remove_wait_queue(&md
->wait
, &wait
);
2228 static void dm_flush(struct mapped_device
*md
)
2230 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
2232 bio_init(&md
->barrier_bio
);
2233 md
->barrier_bio
.bi_bdev
= md
->bdev
;
2234 md
->barrier_bio
.bi_rw
= WRITE_BARRIER
;
2235 __split_and_process_bio(md
, &md
->barrier_bio
);
2237 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
2240 static void process_barrier(struct mapped_device
*md
, struct bio
*bio
)
2242 md
->barrier_error
= 0;
2246 if (!bio_empty_barrier(bio
)) {
2247 __split_and_process_bio(md
, bio
);
2251 if (md
->barrier_error
!= DM_ENDIO_REQUEUE
)
2252 bio_endio(bio
, md
->barrier_error
);
2254 spin_lock_irq(&md
->deferred_lock
);
2255 bio_list_add_head(&md
->deferred
, bio
);
2256 spin_unlock_irq(&md
->deferred_lock
);
2261 * Process the deferred bios
2263 static void dm_wq_work(struct work_struct
*work
)
2265 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2269 down_write(&md
->io_lock
);
2271 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2272 spin_lock_irq(&md
->deferred_lock
);
2273 c
= bio_list_pop(&md
->deferred
);
2274 spin_unlock_irq(&md
->deferred_lock
);
2277 clear_bit(DMF_QUEUE_IO_TO_THREAD
, &md
->flags
);
2281 up_write(&md
->io_lock
);
2283 if (dm_request_based(md
))
2284 generic_make_request(c
);
2286 if (bio_rw_flagged(c
, BIO_RW_BARRIER
))
2287 process_barrier(md
, c
);
2289 __split_and_process_bio(md
, c
);
2292 down_write(&md
->io_lock
);
2295 up_write(&md
->io_lock
);
2298 static void dm_queue_flush(struct mapped_device
*md
)
2300 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2301 smp_mb__after_clear_bit();
2302 queue_work(md
->wq
, &md
->work
);
2305 static void dm_rq_set_flush_nr(struct request
*clone
, unsigned flush_nr
)
2307 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
2309 tio
->info
.flush_request
= flush_nr
;
2312 /* Issue barrier requests to targets and wait for their completion. */
2313 static int dm_rq_barrier(struct mapped_device
*md
)
2316 struct dm_table
*map
= dm_get_live_table(md
);
2317 unsigned num_targets
= dm_table_get_num_targets(map
);
2318 struct dm_target
*ti
;
2319 struct request
*clone
;
2321 md
->barrier_error
= 0;
2323 for (i
= 0; i
< num_targets
; i
++) {
2324 ti
= dm_table_get_target(map
, i
);
2325 for (j
= 0; j
< ti
->num_flush_requests
; j
++) {
2326 clone
= clone_rq(md
->flush_request
, md
, GFP_NOIO
);
2327 dm_rq_set_flush_nr(clone
, j
);
2328 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
2329 map_request(ti
, clone
, md
);
2333 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
2336 return md
->barrier_error
;
2339 static void dm_rq_barrier_work(struct work_struct
*work
)
2342 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2344 struct request_queue
*q
= md
->queue
;
2346 unsigned long flags
;
2349 * Hold the md reference here and leave it at the last part so that
2350 * the md can't be deleted by device opener when the barrier request
2355 error
= dm_rq_barrier(md
);
2357 rq
= md
->flush_request
;
2358 md
->flush_request
= NULL
;
2360 if (error
== DM_ENDIO_REQUEUE
) {
2361 spin_lock_irqsave(q
->queue_lock
, flags
);
2362 blk_requeue_request(q
, rq
);
2363 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2365 blk_end_request_all(rq
, error
);
2373 * Swap in a new table, returning the old one for the caller to destroy.
2375 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2377 struct dm_table
*map
= ERR_PTR(-EINVAL
);
2378 struct queue_limits limits
;
2381 mutex_lock(&md
->suspend_lock
);
2383 /* device must be suspended */
2384 if (!dm_suspended_md(md
))
2387 r
= dm_calculate_queue_limits(table
, &limits
);
2393 /* cannot change the device type, once a table is bound */
2395 (dm_table_get_type(md
->map
) != dm_table_get_type(table
))) {
2396 DMWARN("can't change the device type after a table is bound");
2400 map
= __bind(md
, table
, &limits
);
2403 mutex_unlock(&md
->suspend_lock
);
2408 * Functions to lock and unlock any filesystem running on the
2411 static int lock_fs(struct mapped_device
*md
)
2415 WARN_ON(md
->frozen_sb
);
2417 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2418 if (IS_ERR(md
->frozen_sb
)) {
2419 r
= PTR_ERR(md
->frozen_sb
);
2420 md
->frozen_sb
= NULL
;
2424 set_bit(DMF_FROZEN
, &md
->flags
);
2429 static void unlock_fs(struct mapped_device
*md
)
2431 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2434 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2435 md
->frozen_sb
= NULL
;
2436 clear_bit(DMF_FROZEN
, &md
->flags
);
2440 * We need to be able to change a mapping table under a mounted
2441 * filesystem. For example we might want to move some data in
2442 * the background. Before the table can be swapped with
2443 * dm_bind_table, dm_suspend must be called to flush any in
2444 * flight bios and ensure that any further io gets deferred.
2447 * Suspend mechanism in request-based dm.
2449 * 1. Flush all I/Os by lock_fs() if needed.
2450 * 2. Stop dispatching any I/O by stopping the request_queue.
2451 * 3. Wait for all in-flight I/Os to be completed or requeued.
2453 * To abort suspend, start the request_queue.
2455 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2457 struct dm_table
*map
= NULL
;
2459 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
2460 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
2462 mutex_lock(&md
->suspend_lock
);
2464 if (dm_suspended_md(md
)) {
2469 map
= dm_get_live_table(md
);
2472 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2473 * This flag is cleared before dm_suspend returns.
2476 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2478 /* This does not get reverted if there's an error later. */
2479 dm_table_presuspend_targets(map
);
2482 * Flush I/O to the device.
2483 * Any I/O submitted after lock_fs() may not be flushed.
2484 * noflush takes precedence over do_lockfs.
2485 * (lock_fs() flushes I/Os and waits for them to complete.)
2487 if (!noflush
&& do_lockfs
) {
2494 * Here we must make sure that no processes are submitting requests
2495 * to target drivers i.e. no one may be executing
2496 * __split_and_process_bio. This is called from dm_request and
2499 * To get all processes out of __split_and_process_bio in dm_request,
2500 * we take the write lock. To prevent any process from reentering
2501 * __split_and_process_bio from dm_request, we set
2502 * DMF_QUEUE_IO_TO_THREAD.
2504 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
2505 * and call flush_workqueue(md->wq). flush_workqueue will wait until
2506 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
2507 * further calls to __split_and_process_bio from dm_wq_work.
2509 down_write(&md
->io_lock
);
2510 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2511 set_bit(DMF_QUEUE_IO_TO_THREAD
, &md
->flags
);
2512 up_write(&md
->io_lock
);
2515 * Request-based dm uses md->wq for barrier (dm_rq_barrier_work) which
2516 * can be kicked until md->queue is stopped. So stop md->queue before
2519 if (dm_request_based(md
))
2520 stop_queue(md
->queue
);
2522 flush_workqueue(md
->wq
);
2525 * At this point no more requests are entering target request routines.
2526 * We call dm_wait_for_completion to wait for all existing requests
2529 r
= dm_wait_for_completion(md
, TASK_INTERRUPTIBLE
);
2531 down_write(&md
->io_lock
);
2533 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2534 up_write(&md
->io_lock
);
2536 /* were we interrupted ? */
2540 if (dm_request_based(md
))
2541 start_queue(md
->queue
);
2544 goto out
; /* pushback list is already flushed, so skip flush */
2548 * If dm_wait_for_completion returned 0, the device is completely
2549 * quiescent now. There is no request-processing activity. All new
2550 * requests are being added to md->deferred list.
2553 set_bit(DMF_SUSPENDED
, &md
->flags
);
2555 dm_table_postsuspend_targets(map
);
2561 mutex_unlock(&md
->suspend_lock
);
2565 int dm_resume(struct mapped_device
*md
)
2568 struct dm_table
*map
= NULL
;
2570 mutex_lock(&md
->suspend_lock
);
2571 if (!dm_suspended_md(md
))
2574 map
= dm_get_live_table(md
);
2575 if (!map
|| !dm_table_get_size(map
))
2578 r
= dm_table_resume_targets(map
);
2585 * Flushing deferred I/Os must be done after targets are resumed
2586 * so that mapping of targets can work correctly.
2587 * Request-based dm is queueing the deferred I/Os in its request_queue.
2589 if (dm_request_based(md
))
2590 start_queue(md
->queue
);
2594 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2596 dm_table_unplug_all(map
);
2600 mutex_unlock(&md
->suspend_lock
);
2605 /*-----------------------------------------------------------------
2606 * Event notification.
2607 *---------------------------------------------------------------*/
2608 void dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2611 char udev_cookie
[DM_COOKIE_LENGTH
];
2612 char *envp
[] = { udev_cookie
, NULL
};
2615 kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2617 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2618 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2619 kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
, action
, envp
);
2623 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2625 return atomic_add_return(1, &md
->uevent_seq
);
2628 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2630 return atomic_read(&md
->event_nr
);
2633 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2635 return wait_event_interruptible(md
->eventq
,
2636 (event_nr
!= atomic_read(&md
->event_nr
)));
2639 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2641 unsigned long flags
;
2643 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2644 list_add(elist
, &md
->uevent_list
);
2645 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2649 * The gendisk is only valid as long as you have a reference
2652 struct gendisk
*dm_disk(struct mapped_device
*md
)
2657 struct kobject
*dm_kobject(struct mapped_device
*md
)
2663 * struct mapped_device should not be exported outside of dm.c
2664 * so use this check to verify that kobj is part of md structure
2666 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2668 struct mapped_device
*md
;
2670 md
= container_of(kobj
, struct mapped_device
, kobj
);
2671 if (&md
->kobj
!= kobj
)
2674 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2682 int dm_suspended_md(struct mapped_device
*md
)
2684 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2687 int dm_suspended(struct dm_target
*ti
)
2689 struct mapped_device
*md
= dm_table_get_md(ti
->table
);
2690 int r
= dm_suspended_md(md
);
2696 EXPORT_SYMBOL_GPL(dm_suspended
);
2698 int dm_noflush_suspending(struct dm_target
*ti
)
2700 struct mapped_device
*md
= dm_table_get_md(ti
->table
);
2701 int r
= __noflush_suspending(md
);
2707 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2709 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
)
2711 struct dm_md_mempools
*pools
= kmalloc(sizeof(*pools
), GFP_KERNEL
);
2716 pools
->io_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2717 mempool_create_slab_pool(MIN_IOS
, _io_cache
) :
2718 mempool_create_slab_pool(MIN_IOS
, _rq_bio_info_cache
);
2719 if (!pools
->io_pool
)
2720 goto free_pools_and_out
;
2722 pools
->tio_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2723 mempool_create_slab_pool(MIN_IOS
, _tio_cache
) :
2724 mempool_create_slab_pool(MIN_IOS
, _rq_tio_cache
);
2725 if (!pools
->tio_pool
)
2726 goto free_io_pool_and_out
;
2728 pools
->bs
= (type
== DM_TYPE_BIO_BASED
) ?
2729 bioset_create(16, 0) : bioset_create(MIN_IOS
, 0);
2731 goto free_tio_pool_and_out
;
2735 free_tio_pool_and_out
:
2736 mempool_destroy(pools
->tio_pool
);
2738 free_io_pool_and_out
:
2739 mempool_destroy(pools
->io_pool
);
2747 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2753 mempool_destroy(pools
->io_pool
);
2755 if (pools
->tio_pool
)
2756 mempool_destroy(pools
->tio_pool
);
2759 bioset_free(pools
->bs
);
2764 static const struct block_device_operations dm_blk_dops
= {
2765 .open
= dm_blk_open
,
2766 .release
= dm_blk_close
,
2767 .ioctl
= dm_blk_ioctl
,
2768 .getgeo
= dm_blk_getgeo
,
2769 .owner
= THIS_MODULE
2772 EXPORT_SYMBOL(dm_get_mapinfo
);
2777 module_init(dm_init
);
2778 module_exit(dm_exit
);
2780 module_param(major
, uint
, 0);
2781 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2782 MODULE_DESCRIPTION(DM_NAME
" driver");
2783 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2784 MODULE_LICENSE("GPL");